Rapid identification of multiplex camellia oil adulteration based on lipidomic fingerprint using laser assisted rapid evaporative ionization mass spectrometry and data fusion combined with machine learning

Camellia oil (CAO) is a high-value edible oil with numerous health benefits; however, its authenticity is often compromised by adulteration with cheaper oils. This study proposes a rapid and robust authenticity analysis method for CAO using laser-assisted rapid evaporative ionization mass spectrometry (LA-REIMS) with complementary analytical techniques and chemometric analysis. Fatty acid composition, attenuated total reflectance Fourier transform infrared spectroscopy spectral fingerprinting, 1 H nuclear magnetic resonance spectroscopy, and color analysis were employed to characterize CAO. Although traditional methods exhibited limitations in detecting low-level adulteration (< 40%), LA-REIMS provided detailed lipidomic fingerprints with minimal sample pretreatment and high throughput. By applying both low- and mid-level data fusion strategies to combine LA-REIMS data with GC and developing eight machine learning classification models, including logistic regression, k-nearest neighbor, support vector machine, decision tree, neural network, Kalman filter, linear discriminant analysis, and random forest (RF), substantial improvements in classification accuracy were achieved. Among these, the RF model, particularly when paired with mid-level data fusion, attained an accuracy of 99.56% in discerning authentic CAO from adulterated samples. These findings demonstrated the feasibility of a digital authenticity testing platform for enhancing food safety and quality control in the edible oil industry. • A LA-REIMS technique was used for authenticity analysis of CAO. • Data fusion combined with ML was suitable for identifying adulterated CAO rapidly. • Data fusion of LA-REIMS and GC data combined with ML enhanced classification accuracy. • Mid-level fusion combined with RF achieved the highest accuracy (99.56%).